[PATCH 00/60] add support for Hopper and Blackwell GPUs

[Ben Skeggs]

On 3/5/25 07:18, Timur Tabi wrote:

> On Fri, 2025-05-02 at 13:56 +1000, Dave Airlie wrote:
>> On Wed, 30 Apr 2025 at 09:49, Ben Skeggs <bskeggs at nvidia.com> wrote:
>>> This series implements support for booting GSP-RM firmware version
>>> 570.144, and adds support for GH100, GB10x, and GB20x GPUs.
>>>
>>> A tree containing this series is available here[1], and a temporary
>>> linux-firmware tree here[2].  Timur will send an official patch to
>>> the linux-firmware tree once this series is closer to being merged.
> FYI:
>
> https://github.com/NVIDIA/linux-firmware/commits/nvidia-staging/

Thanks for this Timur!  And for your help in testing/debugging the 
series in general.

>
>> I've dug through this and been running in on/off over the past few weeks,
>>
>> Overall I don't think this makes things much worse, if we had to keep
>> adding fws I suspect this might get painful.
> Agreed, but we need r570 in order to support Hopper and Blackwell.  r535 just doesn't support those
> GPUs.
>
> The good news is that Nouveau extends support to all GPUs with r570, so distros can replace
> r535 with r570 completely[1].  In addition, Nova-Core will also use r570, so we won't need to
> support a newer firmware version until whatever GPU comes after Blackwell.
>
>
> [1] Apparently there's a small Turing bug with r570, and if that can't get fixed in time, then
> Turing may need to stay on r535.

I found the cause/fix for this yesterday (we were missing our half of a 
hw bug workaround for tu10x, introduced in r555).  I will include it in 
a v2 after I've heard from Danilo.

Ben.